Reprocess failed retraces

Retraces fail. That's life. Sometimes this because of a transient problem, like an invalid sources.list (bug 1044362), and we should reprocess those when the problem is resolved.

This needs the following to notify clients to send a new core dump for failed retraces:

- Remove the address signature part of failed: from the crash_signature_for_stacktrace_address_signature row in the Indexes CF.

This needs the following to clean up failed retraces that just succeeded:

- Checks for a matching failed: row in Buckets when creating a new row.
- Change lp:~ev/oops-repository/whoopsie-daisy bucket() to use the timestamp off the Date field for the OOPS row in question. It current just uses gmtime() and takes the day.
- Gets the failed: row from Buckets and inserts all the OOPS IDs from it into the new Bucket row by calling bucket() for each.
- Removes the row from Buckets

We might want to clean up the counts to factor out the cleared failed bucket:

- clean_up_buckets_function() that:
- Uses get_range on DayBuckets with "failed:" as the start and "failed::" as the end. This would look up a column in Bucket for each found bucket ID and if not found call unbucket().
- unbucket() function that:
  - Uses get_range on DayBuckets with a static bucket ID to find all the relevant rows, then delete them.
  - Uses get_range and remove_counter to clear the counter column for the bucket ID in DayBucketsCount.